xref: /dpdk/examples/ip_fragmentation/main.c (revision 7fe90a66c718665e819ba68396e016ca51778d23)
1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
5  *   All rights reserved.
6  *
7  *   Redistribution and use in source and binary forms, with or without
8  *   modification, are permitted provided that the following conditions
9  *   are met:
10  *
11  *     * Redistributions of source code must retain the above copyright
12  *       notice, this list of conditions and the following disclaimer.
13  *     * Redistributions in binary form must reproduce the above copyright
14  *       notice, this list of conditions and the following disclaimer in
15  *       the documentation and/or other materials provided with the
16  *       distribution.
17  *     * Neither the name of Intel Corporation nor the names of its
18  *       contributors may be used to endorse or promote products derived
19  *       from this software without specific prior written permission.
20  *
21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32  */
33 
34 #include <stdio.h>
35 #include <stdlib.h>
36 #include <stdint.h>
37 #include <inttypes.h>
38 #include <sys/types.h>
39 #include <sys/param.h>
40 #include <string.h>
41 #include <sys/queue.h>
42 #include <stdarg.h>
43 #include <errno.h>
44 #include <getopt.h>
45 
46 #include <rte_common.h>
47 #include <rte_byteorder.h>
48 #include <rte_log.h>
49 #include <rte_memory.h>
50 #include <rte_memcpy.h>
51 #include <rte_memzone.h>
52 #include <rte_eal.h>
53 #include <rte_launch.h>
54 #include <rte_atomic.h>
55 #include <rte_cycles.h>
56 #include <rte_prefetch.h>
57 #include <rte_lcore.h>
58 #include <rte_per_lcore.h>
59 #include <rte_branch_prediction.h>
60 #include <rte_interrupts.h>
61 #include <rte_pci.h>
62 #include <rte_random.h>
63 #include <rte_debug.h>
64 #include <rte_ether.h>
65 #include <rte_ethdev.h>
66 #include <rte_mempool.h>
67 #include <rte_mbuf.h>
68 #include <rte_lpm.h>
69 #include <rte_lpm6.h>
70 #include <rte_ip.h>
71 #include <rte_string_fns.h>
72 
73 #include <rte_ip_frag.h>
74 
75 #define RTE_LOGTYPE_IP_FRAG RTE_LOGTYPE_USER1
76 
77 /* allow max jumbo frame 9.5 KB */
78 #define JUMBO_FRAME_MAX_SIZE	0x2600
79 
80 #define	ROUNDUP_DIV(a, b)	(((a) + (b) - 1) / (b))
81 
82 /*
83  * Default byte size for the IPv6 Maximum Transfer Unit (MTU).
84  * This value includes the size of IPv6 header.
85  */
86 #define	IPV4_MTU_DEFAULT	ETHER_MTU
87 #define	IPV6_MTU_DEFAULT	ETHER_MTU
88 
89 /*
90  * Default payload in bytes for the IPv6 packet.
91  */
92 #define	IPV4_DEFAULT_PAYLOAD	(IPV4_MTU_DEFAULT - sizeof(struct ipv4_hdr))
93 #define	IPV6_DEFAULT_PAYLOAD	(IPV6_MTU_DEFAULT - sizeof(struct ipv6_hdr))
94 
95 /*
96  * Max number of fragments per packet expected - defined by config file.
97  */
98 #define	MAX_PACKET_FRAG RTE_LIBRTE_IP_FRAG_MAX_FRAG
99 
100 #define NB_MBUF   8192
101 
102 #define MAX_PKT_BURST	32
103 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
104 
105 /* Configure how many packets ahead to prefetch, when reading packets */
106 #define PREFETCH_OFFSET	3
107 
108 /*
109  * Configurable number of RX/TX ring descriptors
110  */
111 #define RTE_TEST_RX_DESC_DEFAULT 128
112 #define RTE_TEST_TX_DESC_DEFAULT 512
113 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
114 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
115 
116 /* ethernet addresses of ports */
117 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
118 
119 #ifndef IPv4_BYTES
120 #define IPv4_BYTES_FMT "%" PRIu8 ".%" PRIu8 ".%" PRIu8 ".%" PRIu8
121 #define IPv4_BYTES(addr) \
122 		(uint8_t) (((addr) >> 24) & 0xFF),\
123 		(uint8_t) (((addr) >> 16) & 0xFF),\
124 		(uint8_t) (((addr) >> 8) & 0xFF),\
125 		(uint8_t) ((addr) & 0xFF)
126 #endif
127 
128 #ifndef IPv6_BYTES
129 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
130                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
131 #define IPv6_BYTES(addr) \
132 	addr[0],  addr[1], addr[2],  addr[3], \
133 	addr[4],  addr[5], addr[6],  addr[7], \
134 	addr[8],  addr[9], addr[10], addr[11],\
135 	addr[12], addr[13],addr[14], addr[15]
136 #endif
137 
138 #define IPV6_ADDR_LEN 16
139 
140 /* mask of enabled ports */
141 static int enabled_port_mask = 0;
142 
143 static int rx_queue_per_lcore = 1;
144 
145 #define MBUF_TABLE_SIZE  (2 * MAX(MAX_PKT_BURST, MAX_PACKET_FRAG))
146 
147 struct mbuf_table {
148 	uint16_t len;
149 	struct rte_mbuf *m_table[MBUF_TABLE_SIZE];
150 };
151 
152 struct rx_queue {
153 	struct rte_mempool *direct_pool;
154 	struct rte_mempool *indirect_pool;
155 	struct rte_lpm *lpm;
156 	struct rte_lpm6 *lpm6;
157 	uint8_t portid;
158 };
159 
160 #define MAX_RX_QUEUE_PER_LCORE 16
161 #define MAX_TX_QUEUE_PER_PORT 16
162 struct lcore_queue_conf {
163 	uint16_t n_rx_queue;
164 	uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
165 	struct rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
166 	struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
167 } __rte_cache_aligned;
168 struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
169 
170 static struct rte_eth_conf port_conf = {
171 	.rxmode = {
172 		.max_rx_pkt_len = JUMBO_FRAME_MAX_SIZE,
173 		.split_hdr_size = 0,
174 		.header_split   = 0, /**< Header Split disabled */
175 		.hw_ip_checksum = 1, /**< IP checksum offload enabled */
176 		.hw_vlan_filter = 0, /**< VLAN filtering disabled */
177 		.jumbo_frame    = 1, /**< Jumbo Frame Support enabled */
178 		.hw_strip_crc   = 1, /**< CRC stripped by hardware */
179 	},
180 	.txmode = {
181 		.mq_mode = ETH_MQ_TX_NONE,
182 	},
183 };
184 
185 /*
186  * IPv4 forwarding table
187  */
188 struct l3fwd_ipv4_route {
189 	uint32_t ip;
190 	uint8_t  depth;
191 	uint8_t  if_out;
192 };
193 
194 struct l3fwd_ipv4_route l3fwd_ipv4_route_array[] = {
195 		{IPv4(100,10,0,0), 16, 0},
196 		{IPv4(100,20,0,0), 16, 1},
197 		{IPv4(100,30,0,0), 16, 2},
198 		{IPv4(100,40,0,0), 16, 3},
199 		{IPv4(100,50,0,0), 16, 4},
200 		{IPv4(100,60,0,0), 16, 5},
201 		{IPv4(100,70,0,0), 16, 6},
202 		{IPv4(100,80,0,0), 16, 7},
203 };
204 
205 /*
206  * IPv6 forwarding table
207  */
208 
209 struct l3fwd_ipv6_route {
210 	uint8_t ip[IPV6_ADDR_LEN];
211 	uint8_t depth;
212 	uint8_t if_out;
213 };
214 
215 static struct l3fwd_ipv6_route l3fwd_ipv6_route_array[] = {
216 	{{1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 0},
217 	{{2,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 1},
218 	{{3,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 2},
219 	{{4,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 3},
220 	{{5,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 4},
221 	{{6,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 5},
222 	{{7,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 6},
223 	{{8,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 7},
224 };
225 
226 #define LPM_MAX_RULES         1024
227 #define LPM6_MAX_RULES         1024
228 #define LPM6_NUMBER_TBL8S (1 << 16)
229 
230 struct rte_lpm6_config lpm6_config = {
231 		.max_rules = LPM6_MAX_RULES,
232 		.number_tbl8s = LPM6_NUMBER_TBL8S,
233 		.flags = 0
234 };
235 
236 static struct rte_mempool *socket_direct_pool[RTE_MAX_NUMA_NODES];
237 static struct rte_mempool *socket_indirect_pool[RTE_MAX_NUMA_NODES];
238 static struct rte_lpm *socket_lpm[RTE_MAX_NUMA_NODES];
239 static struct rte_lpm6 *socket_lpm6[RTE_MAX_NUMA_NODES];
240 
241 /* Send burst of packets on an output interface */
242 static inline int
243 send_burst(struct lcore_queue_conf *qconf, uint16_t n, uint8_t port)
244 {
245 	struct rte_mbuf **m_table;
246 	int ret;
247 	uint16_t queueid;
248 
249 	queueid = qconf->tx_queue_id[port];
250 	m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
251 
252 	ret = rte_eth_tx_burst(port, queueid, m_table, n);
253 	if (unlikely(ret < n)) {
254 		do {
255 			rte_pktmbuf_free(m_table[ret]);
256 		} while (++ret < n);
257 	}
258 
259 	return 0;
260 }
261 
262 static inline void
263 l3fwd_simple_forward(struct rte_mbuf *m, struct lcore_queue_conf *qconf,
264 		uint8_t queueid, uint8_t port_in)
265 {
266 	struct rx_queue *rxq;
267 	uint32_t i, len, next_hop;
268 	uint8_t port_out, ipv6;
269 	int32_t len2;
270 
271 	ipv6 = 0;
272 	rxq = &qconf->rx_queue_list[queueid];
273 
274 	/* by default, send everything back to the source port */
275 	port_out = port_in;
276 
277 	/* Remove the Ethernet header and trailer from the input packet */
278 	rte_pktmbuf_adj(m, (uint16_t)sizeof(struct ether_hdr));
279 
280 	/* Build transmission burst */
281 	len = qconf->tx_mbufs[port_out].len;
282 
283 	/* if this is an IPv4 packet */
284 	if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
285 		struct ipv4_hdr *ip_hdr;
286 		uint32_t ip_dst;
287 		/* Read the lookup key (i.e. ip_dst) from the input packet */
288 		ip_hdr = rte_pktmbuf_mtod(m, struct ipv4_hdr *);
289 		ip_dst = rte_be_to_cpu_32(ip_hdr->dst_addr);
290 
291 		/* Find destination port */
292 		if (rte_lpm_lookup(rxq->lpm, ip_dst, &next_hop) == 0 &&
293 				(enabled_port_mask & 1 << next_hop) != 0) {
294 			port_out = next_hop;
295 
296 			/* Build transmission burst for new port */
297 			len = qconf->tx_mbufs[port_out].len;
298 		}
299 
300 		/* if we don't need to do any fragmentation */
301 		if (likely (IPV4_MTU_DEFAULT >= m->pkt_len)) {
302 			qconf->tx_mbufs[port_out].m_table[len] = m;
303 			len2 = 1;
304 		} else {
305 			len2 = rte_ipv4_fragment_packet(m,
306 				&qconf->tx_mbufs[port_out].m_table[len],
307 				(uint16_t)(MBUF_TABLE_SIZE - len),
308 				IPV4_MTU_DEFAULT,
309 				rxq->direct_pool, rxq->indirect_pool);
310 
311 			/* Free input packet */
312 			rte_pktmbuf_free(m);
313 
314 			/* If we fail to fragment the packet */
315 			if (unlikely (len2 < 0))
316 				return;
317 		}
318 	} else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
319 		/* if this is an IPv6 packet */
320 		struct ipv6_hdr *ip_hdr;
321 
322 		ipv6 = 1;
323 
324 		/* Read the lookup key (i.e. ip_dst) from the input packet */
325 		ip_hdr = rte_pktmbuf_mtod(m, struct ipv6_hdr *);
326 
327 		/* Find destination port */
328 		if (rte_lpm6_lookup(rxq->lpm6, ip_hdr->dst_addr,
329 						&next_hop) == 0 &&
330 				(enabled_port_mask & 1 << next_hop) != 0) {
331 			port_out = next_hop;
332 
333 			/* Build transmission burst for new port */
334 			len = qconf->tx_mbufs[port_out].len;
335 		}
336 
337 		/* if we don't need to do any fragmentation */
338 		if (likely (IPV6_MTU_DEFAULT >= m->pkt_len)) {
339 			qconf->tx_mbufs[port_out].m_table[len] = m;
340 			len2 = 1;
341 		} else {
342 			len2 = rte_ipv6_fragment_packet(m,
343 				&qconf->tx_mbufs[port_out].m_table[len],
344 				(uint16_t)(MBUF_TABLE_SIZE - len),
345 				IPV6_MTU_DEFAULT,
346 				rxq->direct_pool, rxq->indirect_pool);
347 
348 			/* Free input packet */
349 			rte_pktmbuf_free(m);
350 
351 			/* If we fail to fragment the packet */
352 			if (unlikely (len2 < 0))
353 				return;
354 		}
355 	}
356 	/* else, just forward the packet */
357 	else {
358 		qconf->tx_mbufs[port_out].m_table[len] = m;
359 		len2 = 1;
360 	}
361 
362 	for (i = len; i < len + len2; i ++) {
363 		void *d_addr_bytes;
364 
365 		m = qconf->tx_mbufs[port_out].m_table[i];
366 		struct ether_hdr *eth_hdr = (struct ether_hdr *)
367 			rte_pktmbuf_prepend(m, (uint16_t)sizeof(struct ether_hdr));
368 		if (eth_hdr == NULL) {
369 			rte_panic("No headroom in mbuf.\n");
370 		}
371 
372 		m->l2_len = sizeof(struct ether_hdr);
373 
374 		/* 02:00:00:00:00:xx */
375 		d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
376 		*((uint64_t *)d_addr_bytes) = 0x000000000002 + ((uint64_t)port_out << 40);
377 
378 		/* src addr */
379 		ether_addr_copy(&ports_eth_addr[port_out], &eth_hdr->s_addr);
380 		if (ipv6)
381 			eth_hdr->ether_type = rte_be_to_cpu_16(ETHER_TYPE_IPv6);
382 		else
383 			eth_hdr->ether_type = rte_be_to_cpu_16(ETHER_TYPE_IPv4);
384 	}
385 
386 	len += len2;
387 
388 	if (likely(len < MAX_PKT_BURST)) {
389 		qconf->tx_mbufs[port_out].len = (uint16_t)len;
390 		return;
391 	}
392 
393 	/* Transmit packets */
394 	send_burst(qconf, (uint16_t)len, port_out);
395 	qconf->tx_mbufs[port_out].len = 0;
396 }
397 
398 /* main processing loop */
399 static int
400 main_loop(__attribute__((unused)) void *dummy)
401 {
402 	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
403 	unsigned lcore_id;
404 	uint64_t prev_tsc, diff_tsc, cur_tsc;
405 	int i, j, nb_rx;
406 	uint8_t portid;
407 	struct lcore_queue_conf *qconf;
408 	const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
409 
410 	prev_tsc = 0;
411 
412 	lcore_id = rte_lcore_id();
413 	qconf = &lcore_queue_conf[lcore_id];
414 
415 	if (qconf->n_rx_queue == 0) {
416 		RTE_LOG(INFO, IP_FRAG, "lcore %u has nothing to do\n", lcore_id);
417 		return 0;
418 	}
419 
420 	RTE_LOG(INFO, IP_FRAG, "entering main loop on lcore %u\n", lcore_id);
421 
422 	for (i = 0; i < qconf->n_rx_queue; i++) {
423 
424 		portid = qconf->rx_queue_list[i].portid;
425 		RTE_LOG(INFO, IP_FRAG, " -- lcoreid=%u portid=%d\n", lcore_id,
426 				(int) portid);
427 	}
428 
429 	while (1) {
430 
431 		cur_tsc = rte_rdtsc();
432 
433 		/*
434 		 * TX burst queue drain
435 		 */
436 		diff_tsc = cur_tsc - prev_tsc;
437 		if (unlikely(diff_tsc > drain_tsc)) {
438 
439 			/*
440 			 * This could be optimized (use queueid instead of
441 			 * portid), but it is not called so often
442 			 */
443 			for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
444 				if (qconf->tx_mbufs[portid].len == 0)
445 					continue;
446 				send_burst(&lcore_queue_conf[lcore_id],
447 					   qconf->tx_mbufs[portid].len,
448 					   portid);
449 				qconf->tx_mbufs[portid].len = 0;
450 			}
451 
452 			prev_tsc = cur_tsc;
453 		}
454 
455 		/*
456 		 * Read packet from RX queues
457 		 */
458 		for (i = 0; i < qconf->n_rx_queue; i++) {
459 
460 			portid = qconf->rx_queue_list[i].portid;
461 			nb_rx = rte_eth_rx_burst(portid, 0, pkts_burst,
462 						 MAX_PKT_BURST);
463 
464 			/* Prefetch first packets */
465 			for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
466 				rte_prefetch0(rte_pktmbuf_mtod(
467 						pkts_burst[j], void *));
468 			}
469 
470 			/* Prefetch and forward already prefetched packets */
471 			for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
472 				rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
473 						j + PREFETCH_OFFSET], void *));
474 				l3fwd_simple_forward(pkts_burst[j], qconf, i, portid);
475 			}
476 
477 			/* Forward remaining prefetched packets */
478 			for (; j < nb_rx; j++) {
479 				l3fwd_simple_forward(pkts_burst[j], qconf, i, portid);
480 			}
481 		}
482 	}
483 }
484 
485 /* display usage */
486 static void
487 print_usage(const char *prgname)
488 {
489 	printf("%s [EAL options] -- -p PORTMASK [-q NQ]\n"
490 	       "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
491 	       "  -q NQ: number of queue (=ports) per lcore (default is 1)\n",
492 	       prgname);
493 }
494 
495 static int
496 parse_portmask(const char *portmask)
497 {
498 	char *end = NULL;
499 	unsigned long pm;
500 
501 	/* parse hexadecimal string */
502 	pm = strtoul(portmask, &end, 16);
503 	if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
504 		return -1;
505 
506 	if (pm == 0)
507 		return -1;
508 
509 	return pm;
510 }
511 
512 static int
513 parse_nqueue(const char *q_arg)
514 {
515 	char *end = NULL;
516 	unsigned long n;
517 
518 	/* parse hexadecimal string */
519 	n = strtoul(q_arg, &end, 10);
520 	if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
521 		return -1;
522 	if (n == 0)
523 		return -1;
524 	if (n >= MAX_RX_QUEUE_PER_LCORE)
525 		return -1;
526 
527 	return n;
528 }
529 
530 /* Parse the argument given in the command line of the application */
531 static int
532 parse_args(int argc, char **argv)
533 {
534 	int opt, ret;
535 	char **argvopt;
536 	int option_index;
537 	char *prgname = argv[0];
538 	static struct option lgopts[] = {
539 		{NULL, 0, 0, 0}
540 	};
541 
542 	argvopt = argv;
543 
544 	while ((opt = getopt_long(argc, argvopt, "p:q:",
545 				  lgopts, &option_index)) != EOF) {
546 
547 		switch (opt) {
548 		/* portmask */
549 		case 'p':
550 			enabled_port_mask = parse_portmask(optarg);
551 			if (enabled_port_mask < 0) {
552 				printf("invalid portmask\n");
553 				print_usage(prgname);
554 				return -1;
555 			}
556 			break;
557 
558 		/* nqueue */
559 		case 'q':
560 			rx_queue_per_lcore = parse_nqueue(optarg);
561 			if (rx_queue_per_lcore < 0) {
562 				printf("invalid queue number\n");
563 				print_usage(prgname);
564 				return -1;
565 			}
566 			break;
567 
568 		/* long options */
569 		case 0:
570 			print_usage(prgname);
571 			return -1;
572 
573 		default:
574 			print_usage(prgname);
575 			return -1;
576 		}
577 	}
578 
579 	if (enabled_port_mask == 0) {
580 		printf("portmask not specified\n");
581 		print_usage(prgname);
582 		return -1;
583 	}
584 
585 	if (optind >= 0)
586 		argv[optind-1] = prgname;
587 
588 	ret = optind-1;
589 	optind = 1; /* reset getopt lib */
590 	return ret;
591 }
592 
593 static void
594 print_ethaddr(const char *name, struct ether_addr *eth_addr)
595 {
596 	char buf[ETHER_ADDR_FMT_SIZE];
597 	ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
598 	printf("%s%s", name, buf);
599 }
600 
601 /* Check the link status of all ports in up to 9s, and print them finally */
602 static void
603 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
604 {
605 #define CHECK_INTERVAL 100 /* 100ms */
606 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
607 	uint8_t portid, count, all_ports_up, print_flag = 0;
608 	struct rte_eth_link link;
609 
610 	printf("\nChecking link status");
611 	fflush(stdout);
612 	for (count = 0; count <= MAX_CHECK_TIME; count++) {
613 		all_ports_up = 1;
614 		for (portid = 0; portid < port_num; portid++) {
615 			if ((port_mask & (1 << portid)) == 0)
616 				continue;
617 			memset(&link, 0, sizeof(link));
618 			rte_eth_link_get_nowait(portid, &link);
619 			/* print link status if flag set */
620 			if (print_flag == 1) {
621 				if (link.link_status)
622 					printf("Port %d Link Up - speed %u "
623 						"Mbps - %s\n", (uint8_t)portid,
624 						(unsigned)link.link_speed,
625 				(link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
626 					("full-duplex") : ("half-duplex\n"));
627 				else
628 					printf("Port %d Link Down\n",
629 							(uint8_t)portid);
630 				continue;
631 			}
632 			/* clear all_ports_up flag if any link down */
633 			if (link.link_status == ETH_LINK_DOWN) {
634 				all_ports_up = 0;
635 				break;
636 			}
637 		}
638 		/* after finally printing all link status, get out */
639 		if (print_flag == 1)
640 			break;
641 
642 		if (all_ports_up == 0) {
643 			printf(".");
644 			fflush(stdout);
645 			rte_delay_ms(CHECK_INTERVAL);
646 		}
647 
648 		/* set the print_flag if all ports up or timeout */
649 		if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
650 			print_flag = 1;
651 			printf("\ndone\n");
652 		}
653 	}
654 }
655 
656 /* Check L3 packet type detection capablity of the NIC port */
657 static int
658 check_ptype(int portid)
659 {
660 	int i, ret;
661 	int ptype_l3_ipv4 = 0, ptype_l3_ipv6 = 0;
662 	uint32_t ptype_mask = RTE_PTYPE_L3_MASK;
663 
664 	ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, NULL, 0);
665 	if (ret <= 0)
666 		return 0;
667 
668 	uint32_t ptypes[ret];
669 
670 	ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, ptypes, ret);
671 	for (i = 0; i < ret; ++i) {
672 		if (ptypes[i] & RTE_PTYPE_L3_IPV4)
673 			ptype_l3_ipv4 = 1;
674 		if (ptypes[i] & RTE_PTYPE_L3_IPV6)
675 			ptype_l3_ipv6 = 1;
676 	}
677 
678 	if (ptype_l3_ipv4 == 0)
679 		printf("port %d cannot parse RTE_PTYPE_L3_IPV4\n", portid);
680 
681 	if (ptype_l3_ipv6 == 0)
682 		printf("port %d cannot parse RTE_PTYPE_L3_IPV6\n", portid);
683 
684 	if (ptype_l3_ipv4 && ptype_l3_ipv6)
685 		return 1;
686 
687 	return 0;
688 
689 }
690 
691 /* Parse packet type of a packet by SW */
692 static inline void
693 parse_ptype(struct rte_mbuf *m)
694 {
695 	struct ether_hdr *eth_hdr;
696 	uint32_t packet_type = RTE_PTYPE_UNKNOWN;
697 	uint16_t ether_type;
698 
699 	eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
700 	ether_type = eth_hdr->ether_type;
701 	if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv4))
702 		packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
703 	else if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv6))
704 		packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
705 
706 	m->packet_type = packet_type;
707 }
708 
709 /* callback function to detect packet type for a queue of a port */
710 static uint16_t
711 cb_parse_ptype(uint8_t port __rte_unused, uint16_t queue __rte_unused,
712 		   struct rte_mbuf *pkts[], uint16_t nb_pkts,
713 		   uint16_t max_pkts __rte_unused,
714 		   void *user_param __rte_unused)
715 {
716 	uint16_t i;
717 
718 	for (i = 0; i < nb_pkts; ++i)
719 		parse_ptype(pkts[i]);
720 
721 	return nb_pkts;
722 }
723 
724 static int
725 init_routing_table(void)
726 {
727 	struct rte_lpm *lpm;
728 	struct rte_lpm6 *lpm6;
729 	int socket, ret;
730 	unsigned i;
731 
732 	for (socket = 0; socket < RTE_MAX_NUMA_NODES; socket++) {
733 		if (socket_lpm[socket]) {
734 			lpm = socket_lpm[socket];
735 			/* populate the LPM table */
736 			for (i = 0; i < RTE_DIM(l3fwd_ipv4_route_array); i++) {
737 				ret = rte_lpm_add(lpm,
738 					l3fwd_ipv4_route_array[i].ip,
739 					l3fwd_ipv4_route_array[i].depth,
740 					l3fwd_ipv4_route_array[i].if_out);
741 
742 				if (ret < 0) {
743 					RTE_LOG(ERR, IP_FRAG, "Unable to add entry %i to the l3fwd "
744 						"LPM table\n", i);
745 					return -1;
746 				}
747 
748 				RTE_LOG(INFO, IP_FRAG, "Socket %i: adding route " IPv4_BYTES_FMT
749 						"/%d (port %d)\n",
750 					socket,
751 					IPv4_BYTES(l3fwd_ipv4_route_array[i].ip),
752 					l3fwd_ipv4_route_array[i].depth,
753 					l3fwd_ipv4_route_array[i].if_out);
754 			}
755 		}
756 
757 		if (socket_lpm6[socket]) {
758 			lpm6 = socket_lpm6[socket];
759 			/* populate the LPM6 table */
760 			for (i = 0; i < RTE_DIM(l3fwd_ipv6_route_array); i++) {
761 				ret = rte_lpm6_add(lpm6,
762 					l3fwd_ipv6_route_array[i].ip,
763 					l3fwd_ipv6_route_array[i].depth,
764 					l3fwd_ipv6_route_array[i].if_out);
765 
766 				if (ret < 0) {
767 					RTE_LOG(ERR, IP_FRAG, "Unable to add entry %i to the l3fwd "
768 						"LPM6 table\n", i);
769 					return -1;
770 				}
771 
772 				RTE_LOG(INFO, IP_FRAG, "Socket %i: adding route " IPv6_BYTES_FMT
773 						"/%d (port %d)\n",
774 					socket,
775 					IPv6_BYTES(l3fwd_ipv6_route_array[i].ip),
776 					l3fwd_ipv6_route_array[i].depth,
777 					l3fwd_ipv6_route_array[i].if_out);
778 			}
779 		}
780 	}
781 	return 0;
782 }
783 
784 static int
785 init_mem(void)
786 {
787 	char buf[PATH_MAX];
788 	struct rte_mempool *mp;
789 	struct rte_lpm *lpm;
790 	struct rte_lpm6 *lpm6;
791 	struct rte_lpm_config lpm_config;
792 	int socket;
793 	unsigned lcore_id;
794 
795 	/* traverse through lcores and initialize structures on each socket */
796 
797 	for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
798 
799 		if (rte_lcore_is_enabled(lcore_id) == 0)
800 			continue;
801 
802 		socket = rte_lcore_to_socket_id(lcore_id);
803 
804 		if (socket == SOCKET_ID_ANY)
805 			socket = 0;
806 
807 		if (socket_direct_pool[socket] == NULL) {
808 			RTE_LOG(INFO, IP_FRAG, "Creating direct mempool on socket %i\n",
809 					socket);
810 			snprintf(buf, sizeof(buf), "pool_direct_%i", socket);
811 
812 			mp = rte_pktmbuf_pool_create(buf, NB_MBUF, 32,
813 				0, RTE_MBUF_DEFAULT_BUF_SIZE, socket);
814 			if (mp == NULL) {
815 				RTE_LOG(ERR, IP_FRAG, "Cannot create direct mempool\n");
816 				return -1;
817 			}
818 			socket_direct_pool[socket] = mp;
819 		}
820 
821 		if (socket_indirect_pool[socket] == NULL) {
822 			RTE_LOG(INFO, IP_FRAG, "Creating indirect mempool on socket %i\n",
823 					socket);
824 			snprintf(buf, sizeof(buf), "pool_indirect_%i", socket);
825 
826 			mp = rte_pktmbuf_pool_create(buf, NB_MBUF, 32, 0, 0,
827 				socket);
828 			if (mp == NULL) {
829 				RTE_LOG(ERR, IP_FRAG, "Cannot create indirect mempool\n");
830 				return -1;
831 			}
832 			socket_indirect_pool[socket] = mp;
833 		}
834 
835 		if (socket_lpm[socket] == NULL) {
836 			RTE_LOG(INFO, IP_FRAG, "Creating LPM table on socket %i\n", socket);
837 			snprintf(buf, sizeof(buf), "IP_FRAG_LPM_%i", socket);
838 
839 			lpm_config.max_rules = LPM_MAX_RULES;
840 			lpm_config.number_tbl8s = 256;
841 			lpm_config.flags = 0;
842 
843 			lpm = rte_lpm_create(buf, socket, &lpm_config);
844 			if (lpm == NULL) {
845 				RTE_LOG(ERR, IP_FRAG, "Cannot create LPM table\n");
846 				return -1;
847 			}
848 			socket_lpm[socket] = lpm;
849 		}
850 
851 		if (socket_lpm6[socket] == NULL) {
852 			RTE_LOG(INFO, IP_FRAG, "Creating LPM6 table on socket %i\n", socket);
853 			snprintf(buf, sizeof(buf), "IP_FRAG_LPM_%i", socket);
854 
855 			lpm6 = rte_lpm6_create(buf, socket, &lpm6_config);
856 			if (lpm6 == NULL) {
857 				RTE_LOG(ERR, IP_FRAG, "Cannot create LPM table\n");
858 				return -1;
859 			}
860 			socket_lpm6[socket] = lpm6;
861 		}
862 	}
863 
864 	return 0;
865 }
866 
867 int
868 main(int argc, char **argv)
869 {
870 	struct lcore_queue_conf *qconf;
871 	struct rte_eth_dev_info dev_info;
872 	struct rte_eth_txconf *txconf;
873 	struct rx_queue *rxq;
874 	int socket, ret;
875 	unsigned nb_ports;
876 	uint16_t queueid = 0;
877 	unsigned lcore_id = 0, rx_lcore_id = 0;
878 	uint32_t n_tx_queue, nb_lcores;
879 	uint8_t portid;
880 
881 	/* init EAL */
882 	ret = rte_eal_init(argc, argv);
883 	if (ret < 0)
884 		rte_exit(EXIT_FAILURE, "rte_eal_init failed");
885 	argc -= ret;
886 	argv += ret;
887 
888 	/* parse application arguments (after the EAL ones) */
889 	ret = parse_args(argc, argv);
890 	if (ret < 0)
891 		rte_exit(EXIT_FAILURE, "Invalid arguments");
892 
893 	nb_ports = rte_eth_dev_count();
894 	if (nb_ports == 0)
895 		rte_exit(EXIT_FAILURE, "No ports found!\n");
896 
897 	nb_lcores = rte_lcore_count();
898 
899 	/* initialize structures (mempools, lpm etc.) */
900 	if (init_mem() < 0)
901 		rte_panic("Cannot initialize memory structures!\n");
902 
903 	/* check if portmask has non-existent ports */
904 	if (enabled_port_mask & ~(RTE_LEN2MASK(nb_ports, unsigned)))
905 		rte_exit(EXIT_FAILURE, "Non-existent ports in portmask!\n");
906 
907 	/* initialize all ports */
908 	for (portid = 0; portid < nb_ports; portid++) {
909 		/* skip ports that are not enabled */
910 		if ((enabled_port_mask & (1 << portid)) == 0) {
911 			printf("Skipping disabled port %d\n", portid);
912 			continue;
913 		}
914 
915 		qconf = &lcore_queue_conf[rx_lcore_id];
916 
917 		/* limit the frame size to the maximum supported by NIC */
918 		rte_eth_dev_info_get(portid, &dev_info);
919 		port_conf.rxmode.max_rx_pkt_len = RTE_MIN(
920 		    dev_info.max_rx_pktlen, port_conf.rxmode.max_rx_pkt_len);
921 
922 		/* get the lcore_id for this port */
923 		while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
924 		       qconf->n_rx_queue == (unsigned)rx_queue_per_lcore) {
925 
926 			rx_lcore_id ++;
927 			if (rx_lcore_id >= RTE_MAX_LCORE)
928 				rte_exit(EXIT_FAILURE, "Not enough cores\n");
929 
930 			qconf = &lcore_queue_conf[rx_lcore_id];
931 		}
932 
933 		socket = (int) rte_lcore_to_socket_id(rx_lcore_id);
934 		if (socket == SOCKET_ID_ANY)
935 			socket = 0;
936 
937 		rxq = &qconf->rx_queue_list[qconf->n_rx_queue];
938 		rxq->portid = portid;
939 		rxq->direct_pool = socket_direct_pool[socket];
940 		rxq->indirect_pool = socket_indirect_pool[socket];
941 		rxq->lpm = socket_lpm[socket];
942 		rxq->lpm6 = socket_lpm6[socket];
943 		qconf->n_rx_queue++;
944 
945 		/* init port */
946 		printf("Initializing port %d on lcore %u...", portid,
947 		       rx_lcore_id);
948 		fflush(stdout);
949 
950 		n_tx_queue = nb_lcores;
951 		if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
952 			n_tx_queue = MAX_TX_QUEUE_PER_PORT;
953 		ret = rte_eth_dev_configure(portid, 1, (uint16_t)n_tx_queue,
954 					    &port_conf);
955 		if (ret < 0) {
956 			printf("\n");
957 			rte_exit(EXIT_FAILURE, "Cannot configure device: "
958 				"err=%d, port=%d\n",
959 				ret, portid);
960 		}
961 
962 		ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
963 					    &nb_txd);
964 		if (ret < 0) {
965 			printf("\n");
966 			rte_exit(EXIT_FAILURE, "Cannot adjust number of "
967 				"descriptors: err=%d, port=%d\n", ret, portid);
968 		}
969 
970 		/* init one RX queue */
971 		ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
972 					     socket, NULL,
973 					     socket_direct_pool[socket]);
974 		if (ret < 0) {
975 			printf("\n");
976 			rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: "
977 				"err=%d, port=%d\n",
978 				ret, portid);
979 		}
980 
981 		rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
982 		print_ethaddr(" Address:", &ports_eth_addr[portid]);
983 		printf("\n");
984 
985 		/* init one TX queue per couple (lcore,port) */
986 		queueid = 0;
987 		for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
988 			if (rte_lcore_is_enabled(lcore_id) == 0)
989 				continue;
990 
991 			socket = (int) rte_lcore_to_socket_id(lcore_id);
992 			printf("txq=%u,%d ", lcore_id, queueid);
993 			fflush(stdout);
994 
995 			txconf = &dev_info.default_txconf;
996 			txconf->txq_flags = 0;
997 			ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
998 						     socket, txconf);
999 			if (ret < 0) {
1000 				printf("\n");
1001 				rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: "
1002 					"err=%d, port=%d\n", ret, portid);
1003 			}
1004 
1005 			qconf = &lcore_queue_conf[lcore_id];
1006 			qconf->tx_queue_id[portid] = queueid;
1007 			queueid++;
1008 		}
1009 
1010 		printf("\n");
1011 	}
1012 
1013 	printf("\n");
1014 
1015 	/* start ports */
1016 	for (portid = 0; portid < nb_ports; portid++) {
1017 		if ((enabled_port_mask & (1 << portid)) == 0) {
1018 			continue;
1019 		}
1020 		/* Start device */
1021 		ret = rte_eth_dev_start(portid);
1022 		if (ret < 0)
1023 			rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
1024 				ret, portid);
1025 
1026 		rte_eth_promiscuous_enable(portid);
1027 
1028 		if (check_ptype(portid) == 0) {
1029 			rte_eth_add_rx_callback(portid, 0, cb_parse_ptype, NULL);
1030 			printf("Add Rx callback function to detect L3 packet type by SW :"
1031 				" port = %d\n", portid);
1032 		}
1033 	}
1034 
1035 	if (init_routing_table() < 0)
1036 		rte_exit(EXIT_FAILURE, "Cannot init routing table\n");
1037 
1038 	check_all_ports_link_status((uint8_t)nb_ports, enabled_port_mask);
1039 
1040 	/* launch per-lcore init on every lcore */
1041 	rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1042 	RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1043 		if (rte_eal_wait_lcore(lcore_id) < 0)
1044 			return -1;
1045 	}
1046 
1047 	return 0;
1048 }
1049